Available online at www.sciencedirect.

com

ScienceDirect
Journal of Interactive Marketing 44 (2018) 43 – 59
www.elsevier.com/locate/intmar

Touchable Apps: Exploring the Usage of Touch Features and

Their Impact on Engagement
Savannah Wei Shi ⁎& Kirthi Kalyanam
Leavey School of Business, Santa Clara University, 500 El Camino Real, Santa Clara, CA 95053, United States of America

Available online 30 August 2018

Abstract

Touch has become an inseparable element of mobile platforms. This study examines the use of different touch features and the impact of these
touch gestures on consumer engagement with a mobile shopping app. We focus on three informational touch features that are common among
shopping apps: touch to zoom in on a page (zoom-page), to view product details (product-view), and to be directed to outside links (open-
webpage). We develop a two-level model that captures (1) consumers' decisions to stay with or leave an app and (2) their use of touch features. Our
main results empirically demonstrate the strong explanatory power of informational touch gestures, especially their dwell time, in consumer app
browsing decisions, whereas navigational touch gestures do not significantly affect app stay likelihood. A longer dwell time and early use of zoom-
page within a session encourage the stay. Moreover, we observe strong synergy and antergy (negative synergy) among these touch gestures. The
cumulative dwell time and temporal progression of touch gestures affect subsequent touch feature usage. Managerially, our results suggest that an
early intervention that encourages the use of zoom-page increases app stay likelihood, and marketers may apply our model to quantify the impact
of such interventions on consumer browsing decisions at the individual level. The results also shed light on how marketers can infer the stage of the
shopping process based on touch gestures (segmentation) and guide consumers through the purchase funnel by promoting the use of zoom-page
and product-view. Lastly, the findings provide insights into how marketers can promote the use of open-webpage, which has the lowest baseline
usage rate yet is crucial for transactions, based on the synergy among touch gestures and through improving the non-native browsing experience.
© 2018

Keywords: Mobile commerce; Touch features; Touch-stream data; App engagement; Non-native experience

Introduction using a mouse, mimicking how consumers intuitively inspect

Mobile commerce has gained significant momentum in
recent years. Forrester Research (2014) predicted that U.S.
mobile commerce would top $293 billion by 2018 (54% of total
U.S. e-commerce sales), up from $42 billion in 2013. The most
apparent difference between mobile devices and other shopping
channels is that navigation and information-gathering are touch
gesture-based: consumers must physically touch their devices to
go through pages within an app, examine specific products, or
complete transactions. Touch gesture-based inputs enable
consumers to use their fingers to “touch” products without
⁎ Corresponding author.
E-mail addresses: wshi@scu.edu (S.W. Shi), kkalyanam@scu.edu
(K. Kalyanam).
products in an offline environment. A variety of touch features
can be performed, including pinch, spread, (double) tap, and
rotate (Villamor, Willis, and Wroblewski 2010). These touch
features can be broadly categorized into two types: navigational
touch, such as scroll and swipe, which does not reveal additional
information that is not presented by default; the other is
informational touch, such as zoom in or open an embedded
link, which retrieves additional, detailed information that is not
presented by default on the current page. In this paper, we
focused on the use of informational touch.
Informational touch data produce a detailed observational
record of how consumers access incremental information on a
page, enabling researchers and practitioners to explore consumer
browsing decisions at a more granular level. With limited real
estate on mobile platforms, the product information presented by

https://doi.org/10.1016/j.intmar.2018.06.001
1094-9968/© 2018
44 S.W. Shi, K. Kalyanam / Journal of Interactive Marketing 44 (2018) 43–59
default in apps is often in an abbreviated form (e.g., a basic image
of the product with a brief description) compared to that on PCs. If
interested, a consumer must apply informational touch gestures to
review detailed information, either to enlarge the presentation
(e.g., zoom in on a page) or to be directed to an inner layer of
information that is not presented by default (e.g., open an
embedded link). While the exact informational touch features
might vary by app, their functionality is consistent—consumers
must take an active action (touch) to access additional
information. For example, consumers can scroll through headlines
on the Wall Street Journal app with navigational touch, yet to read
an article, they need to use informational touch (to expand the
headline); consumers of the popular WhatsApp can browse the
chat feed using navigational touch but to dive into a particular
post, informational touch is needed. In contrast, it is challenging to
disambiguate from traditional clickstream data whether a
consumer is reading detailed product information or merely
skimming through the abbreviated information (such as an image)
on a page, because more detailed product information is presented
by default given the abundant webpage real estate on a PC.
Informational touch data record a greater level of granularity in the
information retrieval process and are therefore of great interest to
both marketing researchers and practitioners.
The investigation of touch features on mobile devices is a
nascent field. The extant research has focused on the adoption
of mobile devices, the formulation of mobile marketing
strategies, and the investigation of consumer attitudes in the
mobile context (e.g., Koenigstorfer and Groeppel-Klein 2012;
Leppäniemi and Karjaluoto 2008; Shankar and
Balasubramanian 2008; Shankar et al. 2010; Tokárová and
Weideman 2013). However, not much attention has been given
to touch gestures and their use on mobile platforms.
Our research fills this gap by exploring the use of different
informational touch features and the impact of various touch
gestures on consumer engagement in a mobile shopping app.
We focus on three informational touch features that are popular
among many shopping apps, namely, touch to zoom in on a
page, to view product details, and to be directed to outside
links. We develop a two-level model to capture (1) consumers'
decisions to stay with or leave an app and (2) their use of
informational touch features. Our unique observational dataset
recorded the consumer use of touch features and their browsing
behavior in a natural mobile shopping environment, which
alleviates typical concerns regarding self-reported measures or
left truncation bias and allows us to examine consumer usage
behaviors with minimum intrusion.
This research extends the effect of touch from the offline
environment to mobile platforms and deepens our understand-
ing of the ways in which consumer shopping behavior is
influenced by interactive technology. Managerially, touch-
stream data allow marketers to obtain a more refined view of
browsing decisions by revealing consumer intent to retrieve
additional information that is not presented by default. First,
our main results empirically demonstrate the strong explanatory
power of informational touch gestures, especially their dwell
time, in consumer app browsing decisions, whereas naviga-
tional touch gestures do not significantly affect app stay
likelihood. Thus, it is crucial for marketers to closely monitor
informational touch gestures when assessing consumer brows-
ing decisions. Second, our results show that an early
intervention that encourages the use of zoom-page features
improves app stay likelihood; thus, marketers can apply our
framework to quantify the impact of such interventions and the
timing thereof on consumer browsing decisions at the
individual level. Third, we observe strong synergy and antergy
(negative synergy) among these touch gestures, as the
cumulative dwell time of touch gestures and their temporal
progression affect subsequent touch use. These results shed
light on how marketers can infer the stage of the shopping
process based on touch gestures (segmentation) and guide
consumers through the purchase funnel by promoting the use of
certain features (such as zoom-page and product-view). Lastly,
we discuss how marketers can promote the open-webpage,
which has the lowest baseline usage rate yet is vital for
transactions, based on the synergy among touch gestures and
through improving the non-native browsing experience.
The paper proceeds as follows. First, we provide background
on the informational touch features of mobile apps and summarize
the relevant literature. Second, we describe the touch-stream data
and present some preliminary evidence. We then describe our
model and present the results. Lastly, we discuss the managerial
implications and future research directions.
Background and Relevant Literature
The Use Patterns of Informational Touch Features
This paper focuses on three informational touch features that
are ubiquitous in mobile apps, namely, zoom-page, product-
view, and open-webpage (illustrated in Appendix 1).
The use of these touch features can be driven by the different
search goals and the purchase intent they reflect. Zoom-page is
recorded when consumers expand their fingers to zoom in on a
page, indicating a transition from an overview of product
offerings to certain item(s) on the page. Since not much
information can be presented on a catalog page given the size
limit on mobile devices; this touch feature is used to present an
image in greater detail (such as the texture of a cloth) or to
enlarge the font for price and brand information. Consumers
who use this touch feature are more likely to be in the “initial
interest” stage, where they are exploring products on a page but
not ready to expend the extra effort to retrieve additional details
beyond the current page. Thus, the adoption of zoom-page is
likely to happen earlier in a session.
The product-view and open-webpage features, however,
reflect a more focused search for the inner-layer information, and
are likely to happen later in a session. With product-view,
consumers touch a product to view the additional product
information, such as reviews, size, and material, that is not
presented by default, through for example, a pop-up window.
With open-webpage, consumers click on an embedded link (in
our case, “Buy on Retailer's Website”) and are redirected to the
retailer's product webpage (typically, the full [non-mobile] site),
either within the app or in a new window outside the app, for
 S.W. Shi, K. Kalyanam / Journal of Interactive Marketing 44 (2018) 43–59
more details or to complete their transactions. This feature is
often provided for a variety of reasons, including development
cost considerations and agreements between app developers and
retailers.1 Consumers who use these two touch features are more
likely to be in a serious browsing mood or at a later stage of the
purchase funnel, because they are willing to expend extra
cognitive effort to retrieve and evaluate additional information.
Serving a similar purpose, these two touch features might create
synergy in the information search. Yet with different browsing
goals and purchase intent, the use of zoom-page is likely to
create antergy with that of product-view and open-webpage.
Therefore, we expect the use of product-view and open-webpage
to reinforce each other, but that the use of these two touch
gestures is negatively correlated with the use of zoom-page.
Moreover, the baseline usage rate (or intrinsic preference) of
these touch features may differ, depending on whether they
generate a non-native browsing experience. In the mobile
context, when consumers consume the content provided by the
focal app or make transactions within the app, it is considered
as a “native” browsing experience. In contrast, a “non-native”
experience is defined as when the browsing or purchasing
activities are conducted via an embedded, external webpage
link within the app (e.g., through a touch gesture, consumers
open the embedded link to an external blog page or YouTube
video, which are not native information of the target app), or
are directed out of the current app to another window (e.g.,
through a touch gesture, consumers are directed to the Internet
browser to open product websites, to an app marketplace, or to
another app). Current mobile platforms do not support multi-
window operation and switching among different windows can
be cumbersome (e.g., Lawrence, Pernici, and Krogstie 2004);
furthermore, connecting to embedded links requires a signifi-
cantly longer loading time (e.g., the webpage must download
about 0.5 MB of extra resources, such as cascading style
sheets, JavaScript, and images), which can be detrimental to
user experience (e.g., DoubleClick 2016; Galletta et al. 2004).
Thus, consumers might incur additional cognitive or time costs
when connecting to an embedded webpage link. However,
touch gestures inducing native experiences, such as zoom-page
and product-view, create less friction and interruptions in the
browsing experience. All else being equal, the usage likelihood
of zoom-page and product-view should be higher than that of
open-webpage. Hence, we expect the baseline usage rate
(usage frequency) of touch gestures associated with a non-
native experience to be lower than that of touch gestures
associated with a native experience.
The Impact of Touch Gestures on App Engagement
The use of touch features can also affect consumers' app
engagement level. Engagement is defined as “customer's
behavioral manifestations that have a brand or firm focus,
beyond purchase, resulting from motivational drivers” (Van
1
Inserting transactional functions may incur an extra fee; for instance, Stripe,
a mobile payment option, charges 2.9% and 30 cents per transaction.
45
Doorn et al. 2010, p. 254); stay/exit decisions, time spent, and
actions taken, can all be viewed as measurements of
engagement. In this study, we focus on app stay/exit decision
as the dependent variable that reflects engagement.
Consumers touch to encode an object's material properties,
such as information about its texture, hardness, and temperature
(Klatzky and Lederman 1992). In an offline environment,
researchers have shown that physical touch tends to increase
consumer response, achieve greater persuasion, and create a
sense of ownership, even without providing useful product-
related information (e.g., Peck and Childers 2003a, b; Peck and
Shu 2009). The positive impact of touch is typically attributed
to the increased level of physical tangibility, mental tangibility,
and specificity tangibility (Coyle and Thorson 2001; Verhagen
et al. 2013). The tactile stimulation generated by this direct
experience with a product can affect product evaluation and
increase perceived ownership (e.g., Peck and Shu 2009; Smith
and Swinyard 1983).
Products become less tangible when sold online because
physical contact between the fingers and the products is eliminated,
as is haptic feedback (e.g., Laroche, Bergeron, and Goutaland
2001). However, we propose that on the mobile platform, touch
gestures can still benefit app engagement by enhancing the
perceived level of product tangibility. Mobile shopping apps
often enable versatile touch gestures to provide rich and interactive
experiences with product presentations; the elimination of the
intermediary mouse mimics product examination in an offline
environment. Indeed, touch on tablets generates an endowment
effect in lab experiments (Brasel and Gips 2014). Furthermore, an
increased level of user satisfaction in gaming (Ullrich 2005) was
observed when additional tactile feedback was introduced.
However, interacting with multi-touch apps on mobile
devices requires greater scale of movement. From an ergo-
nomic perspective, touch on mobile devices could be effortful
and costly. Kolb, Rudner, and Reichert (2012) suggested that
frequent movement of the whole arm and constantly holding
the arm in the air when interacting with mobile devices can
create fatigue. Heo and Lee (2011) proposed that the backward/
forward motions and thumbing through pages often leads to
more tiredness. From cognitive-effort perspective, product-
view and open-webpage involve searching and evaluating extra
product details (even more so for open-webpage), which are
more cognitively costly, whereas zoom-page only enlarges the
information presented on the current page, which is less
cognitively demanding. In this paper, we assess the impact of
the dwell time of these touch gestures in addition to their
frequency of use, given that the amount of time spent on stimuli
has been positively correlated with cognitive processing load
(He and McCarley 2010; Peterson and Wong 2008;
Underwood, Jebbett, and Roberts 2004).
In sum, touch gestures have the potential for both positive
and negative impacts on mobile app stay/exit decisions. We
therefore expect that these informational touch features to be
predictive of app engagement; and the touch features that
reveal additional, inner-layer information would exhibit a
stronger negative effect on app engagement than those touch
features that do not.
46 S.W. Shi, K. Kalyanam / Journal of Interactive Marketing 44 (2018) 43–59

Table 1 Percentage of Sessions Observing the First-time Usage in Each Quarter

Summary statistics of the variables. 50%

Variables Mean SD
40%
Num. sessions per user 7.81 13.27
Num. navigational swipe per session 83.32 163.55 30%

Perc.
Num. informational swipe per session 3.25 13.08
Num. zoom-page per session 2.10 6.09
20%
Num. product-view per session 1.12 10.94
Num. open-webpage per session 0.04 0.34
10%
Dwell time for navigational touches (in s) 5.59 25.17 1 2 3 4
Quarter
Dwell time of zoom-page (in s) 9.64 15.66
Zoom-Page Prod-View Open-Webpage
Dwell time of product-view (in s) 11.64 18.81
Perc. Sessions Observing the First-time usage in Each Quarter
Dwell time of open-webpage (in s) 9.47 17.01
Quarter Zoom-Page Prod-View Open-Webpage
1 45.99% 37.86% 21.38%
2 26.86% 20.82% 26.04%
Data and Preliminary Evidence 3 16.68% 24.25% 28.62%
4 10.47% 17.07% 23.96%

Our observational data were collected unobtrusively from a Frequency of Usage in Each Quarter Weighted by Total Feature Usage
free mobile shopping app, which aggregates catalogs from 30%
multiple leading retailers, since its initial launch. Consumers 28%
touch their tablets to examine catalogs and products, and when
26%
ready to make a purchase, they use touch to be connected to a
24%
retailers' embedded webpage link within the app to finish
Perc.
transaction. We randomly sampled 4,284 consumers, and 22%

65,107 shopping sessions, for our analyses. Table 1 presents 20%

some descriptive statistics. On average, the consumers visited 18%
the target app 7.81 times (SD = 13.27); at each session, they 1 2
Quarter
3 4

took 83.32 (SD = 163.55) navigational swipes and 3.25 (SD  Zoom-Page
= 13.08) informational swipes. Despite considerable variation
across sessions, the use frequency of the zoom-page was the
highest (2.10, SD = 6.09), followed by product-view (1.12,
SD = 10.94). The use of open-webpage had the lowest value
among the three (0.04, SD = 0.34). The duration of informa-
tional touch features varied as well: the consumers dwelled
about 9.64 s (SD = 15.66) after zooming in on a page and
9.47 s (SD = 17.01) after opening a webpage link; however,
they tended to spend more time after performing product-views
(11.64, SD = 18.81).
Next, we explore when consumers would start using these
informational touch features in a session. To examine the
average progress of touch gestures across all consumers, we
divide total browsing duration of each consumer into four
quarters and examine when consumers started using certain
touch feature.2 Fig. 1.1 presents the percentages of sessions
with the first-time use of three touch features in each quarter;
for zoom-page, 45.99% sessions saw first-time use in the 1st
quarter, and the values decreases monotonically from the 1st to
the 4th quarters (45.99%, 26.86%, 16.68%, and 10.47%,
respectively). This pattern shows that consumers were more
likely to start using zoom-page earlier rather than later in a
session. In contrast, consumers were more likely to start using
open-webpage later in a browsing session, as the percentage of
2
We expect that a user will go through several stages (e.g., initial exploration,
more focused search) during a browsing session, regardless of session length.
Using the quarterly-division (as a relative measure) takes into account the
heterogeneity of individual browsing duration both across and within users.
Using the absolute value of the time may not capture the progress of the
browsing while account for customer heterogeneity in session length.
Prod-View Open-Webpage
Usage Freq. in Each Quarter Weighted by Total Usage Freq
Quarter Zoom-Page Prod-View Open-Webpage
1 23.78% 23.31% 19.76%
2 27.64% 25.78% 26.03%
3 26.00% 26.66% 29.71%
4 22.58% 24.25% 24.50%
Fig. 1. Temporal progression of informational touch features. 1.1. Percentage of
sessions observing the first-time usage in each quarter. 1.2. Frequency of usage
in each quarter weighted by total feature usage.
sessions that saw the first-time usage peaked in the 3rd quarter
(28.62%), which is over 30%increase from its lowest value in
the 1st quarter. The percentage of sessions that saw the first-
time use of product-view did not exhibit a clear trend across the
four quarters, with the highest value observed in the 1st quarter
(37.86%), followed by the 3rd (24.25%), 2nd (20.82%), and 4th
quarters (17.07%). In general, it took an average of 200.50 s
(SD = 268.40) for a consumer to adopt zoom-page, whereas
the average for product-view was 314.90 s (SD = 426.90).
Consumers tended to use open-webpage much later in a
session: it took them over seven minutes to perform this touch
gesture (431.5 s, SD = 516.41).
We next explore how the use frequency share of these touch
features progressed over time in each session. To make the
values comparable, we divide the use frequency of each touch
feature in each quarter for a given consumer by the total use
frequency of that touch feature during the session for that
consumer. As shown in Fig. 1.2, the use of zoom-page was the
most frequent in the 2nd quarter (27.64%) and the least frequent
in the 4th quarter (22.58%). Yet, the use of product-view and
open-webpage was the most frequent in the 3rd quarter
(26.66% and 29.71%, respectively) and the least frequent
 S.W. Shi, K. Kalyanam / Journal of Interactive Marketing 44 (2018) 43–59
during the 1st quarter (23.31% and 19.76% for product-view
and open-webpage, respectively). The above analyses regard-
ing the temporal progression of touch features provide support
to our conjecture that zoom-page represents initial interest and
is often used earlier in a session; while product-view and open-
webpage indicate more focused, goal-directed searches and are
often applied later in a session.
To motivate our model and shed further insights into its
findings, we conducted a brief survey regarding the use of these
three touch features. The details of this survey and its results are
presented in Appendix 2. The participant responses led us to
believe that these touch features are learned quickly, and that
their use is mainly driven by the usage experience with the
target app. Next, we discuss the model formulation.
Model
We formulate consumer browsing decisions and their use of
touch features in a two-step process with the nested logit model
(Train 2009). First, the consumers decide whether to stay or
exit the app. This is a level-one binary decision (the branch
level in the nested logit model). Second, if they decide to stay
with the app, they either proceed to a new page or look for
additional, detailed information on the current page using the
three informational touch features (i.e., zoom-page, product-
view, and open-webpage). This is a level-two multinomial
decision (the twig level in the nested logit model). The
inclusive value of the alternatives in level two (the twig level)
entered the utility in level one (the branch level). This setup
allows us to examine the impact of informational touch features
on app engagement, as well as the drivers behind the use of
different touch features, through a marginal benefit perspective.
Level 1: App Stay Vs. Exit Decisions
Let Iit represents app stay/exit dummy, with Iit = 1
indicating continued visit, and 0 indicating an exit from the
app. We propose that the utility of staying with the app for
consumer i at time t, U1it, would be influenced by within-
session experience, across-session experience, control vari-
ables, and the inclusive value from the alternatives in level-two
choices. The equation for the utility of staying with the app is:
U1it ¼ V 1it þ ε1it
¼ θi  X it þ ε1it
X it : fWithinSessionExperienceit ; AcrossSessionExperienceis ;
ControlVaris ; InclusiveValueg ð1Þ
Parameters θ.i are allowed to vary across individuals to
capture heterogeneity.
The probability of staying with the app is:
expðV 1it Þ
P1ðI it ¼ 1Þ ¼
1 þ expðV 1it Þ
For within-session experience, we examine how a change in
consumers' moment-to-moment experience levels during an app
47
visit influences the decision to stay or exit. The variables include
cumulative time spent (CumuTimeSession, log-transformed) and
the cumulative number of navigational touch (swipe) gestures
(CumuSwipeSession, log-transformed) until time t − 1. To explore
the effects of the temporal progression of three informational touch
features on app engagement, we include the cumulative number of
different (unique) informational touch gestures (CumuType
TouchSession, log-transformed) until time t − 1, the cumulative
dwell time (log-transformed) when performing zoom-page
(CumuZPTimeSession), product-view (CumuPVTimeSession),
and open-webpage (CumuOWTimeSession) until time t − 1.
Several dummy variables are included to describe the
quarter during which we observe the first-time use of these
informational touch gestures. Specifically, we first calculate the
median value of browsing duration of all prior sessions
(sessions 1 to s − 1) for a consumer and then identify the
time points t1, t2, and t3 for the quarter division. Then, based
on these time points, we divide session s into four quarters and
then code for whether the first-time use of a touch feature
occurs in the 1st, 2nd, 3rd, or 4th quarters. The dummy
variables include 1stQuarterObserveFirstZC (whether zoom-page
is first used during the 1st quarter), 2ndQuarterObserveFirstZC
(whether zoom-page is first used during the 2nd quarter),
3rdQuarterObserveFirstZC (whether zoom-page is first used
during the 3rd quarter), 1stQuarterObserveFirstPV (product-
view), 2ndQuarterObserveFirstPV, 3rdQuarterObserveFirstPV,
1stQuarterObserveFirstOW (open-webpage), 2ndQuarter
ObserveFirstOW, and 3rdQuarterObserveFirstOW (the
4th quarter is used as baseline). To summarize, for consumer
i at time t in a session, the within-session experience variables are:
WithinSessionExperienceit ¼
fCumuDuraSessionit ; CumuSwipeSessionit ;
CumuTypeTouchSessionit ; CumuZPTimeSessionit ;
CumuPVTimeSessionit ; CumuOWTimeSessionit ;
1stQuarterObserveFirstZC is ; 2ndQuarterObserveFirstZC is ;
3rdQuarterObserveFirstZC is ; 1stQuarterObserveFirstPV is ;
2ndQuarterObserveFirstPV is ; 3rdQuarterObserveFirstZC is ;
1stQuarterObserveFirstOW is ; 2ndQuarterObserveFirstOW is ;
3rdQuarterObserveFirstOW is g ð3Þ
For the across-session experience, we gauge the impact of
consumers' total experience levels with the target app on their
browsing decisions. The variables include the cumulative time
spent with the current app (CumuDuraApp, log-transformed),
the cumulative number of navigational touch (swipe) gestures
(CumuSwipeApp, log-transformed), and the cumulative number
of informational touch gestures (CumuTouchApp, log-trans-
formed) with the current app until session s − 1. Note that
these variables will influence the baseline of the app browsing
decisions. To summarize, for consumer i at session s, the
ð2Þ
48 S.W. Shi, K. Kalyanam / Journal of Interactive Marketing 44 (2018) 43–59
across-session experience variables are.
AcrossSessionExperienceis
¼ fCumuDuraAppis ; CumuSwipeAppis ; CumuTouchAppis g
ð4Þ
We also control for minor and major app upgrades
performed since the last session, holidays (equals 1 if the
session was during the week of U.S. Thanksgiving or
Christmas, and 0 otherwise), and weekends (equals 1 if the
session was during a weekend; and 0 otherwise, given that the
use of a tablet, a more personal device, often peaks over the
weekend (Salesforce 2014). To summarize, for consumer i at
session s, the control variables are:
ControlVaris
¼ fMinorUpgradeis ; MajorUpgradeis ; Weekend is ; Holidayis g
ð5Þ
Level 2: Touch Gesture and New Page Decisions
Conditional on staying with the app (level 1, Iit = 1),
consumers decide whether to proceed to the next page or look
for additional information on the current page using zoom-page,
product-view, or open-webpage. Assuming that moving to a new
page is the baseline, we propose that the utility of zoom-page,
product-view, and open-webpage are affected by the within-
session experience, across-session experience, page-level dynam-
ics, and control variables. Because there are no choice-varying
variables, we apply a standard multinomial logit choice model in
which the parameters vary across choice (Powers and Xie 2008, p.
246; Train 2009). The utility of choosing touch option j, U2ijt, (j 
= 1, zoom-page; j = 2: product-view; j = 3: open-webpage;
with a new page as baseline) for consumer i at time t is:
U2ijt ¼ V 2ijt þ ε2ijt
¼ β i; j  Z it þ ε2ijt
ð6Þ
Z it : fWithinSessionExperienceit ;AcrossSessionExperienceis ;
PageLevelDynamicsit ; ControlVaris g
Parameters β⋅,i,j vary across options and individuals. The
choice probability of observed touch gesture Tit = j (j = 1,
zoom-page; j = 2: product-view; j = 3: open-webpage) is:


The inclusive value based on level-two choices, to be
inserted in the level-one model, can be written as:
Inclusive Valueit = (ln(∑j ∈ {1, 2, 3} exp (V2ijt/λ) + 1)).
Within-session experience will prompt consumers to use
different touch features. The set of variables in Within-session
experience for the level-two choice are described in Eq. (3).
Across-session experience could alter the baseline usage rate of
these touch features. We resort to the same set of variables
described in Eq. (4).
At the page-level, the use of one informational touch feature
may influence the use of the others: after performing a touch
gesture on a (catalog) page, a consumer may or may not want to
retrieve more detailed, inner-layer information on the same
page. Thus, we use the cumulative dwell time (log-trans-
formed) of zoom-page (CumuZPTimePage), page-view
(CumuPVTimePage), and open-webpage (CumuOWTimePage)
on the current page to capture the information retrieval and
processing behavior on a page. To summarize, for consumer i at
time t in a session, PageLevelDynamicsit are:
PageLevelDynamicsit
¼ fCumuZPTimePageit ; CumuPVTimePageit ; CumuOWTimePageit g
ð9Þ
Similarly, we control for minor or major app upgrades that
have happened since the last session, weekend, or holiday
(variables described in Eq. (5)).
The joint distribution of a consumer's two-level browsing
behavior can be written as:


PrðT it ; I it Þ ¼ P2 T ijt jI it ¼ 1  P1ðI it Þ
ð10Þ
¼ P1ðI it ¼ 0ÞI it ¼0  f½P2ðT it jI it ¼ 1Þ  P1ðI it ¼ 1ÞgI it ¼1
We test our model against two benchmark models. One is a
browsing model without any informational touch predictors. The
variables entering the utility function will be CumuDuraSessionit,
CumuSwipeSessionit, CumuDuraAppis, CumuSwipeAppis,
MinorUpgradeis, MajorUpgradeis, Weekendis, and Holidayis. The
other is a model that used the frequency of informational
touch gestures instead of dwell time of informational touch
gestures. Specifically, we replace the CumuZPTimeSession,
CumuPVTimeSession, CumuOWTimeSession, ZPDwellTimePage,
PVDwellTimePage, and OWDwellTimePage with the frequency
counts of these touch gestures.

exp V 2ijt =λ
P2 T it ¼ jI ðstayÞit ¼ 1 ¼
ð7Þ Model Results
1 þ ∑ j∈f1;2;3g exp V 2ijt =λ

with λ (log-sum coefficient) indicating the degree of indepen- The model was implemented using the Metropolis–Hastings
dence among the unobserved portions of utility for alternatives Bayesian approach with weakly informative priors for all
(three touch gestures) in the “nest” (Train 2009). model parameters. We used 50,000 draws and discarded the
The choice probability of observing new page viewing (Tit  first 5,000 draws to burn-in. Convergence was checked with
= 4) is: standard methods and was achieved well before the end of the
burn-in.

1 The log-likelihood, AIC and BIC values, show that our main
P2 T it −4I ðstayÞit ¼ 1 ¼
ð8Þ model performs better than the model without informational
1 þ ∑ j∈f1;2;3g exp V 2ijt =λ
touch predictors and the model with frequency count instead of
dwell time (Table 2). These comparison statistics demonstrate
 S.W. Shi, K. Kalyanam / Journal of Interactive Marketing 44 (2018) 43–59
Table 2
Model comparison.
Log likelihood AIC BIC
Main model − 224,783.50 449,771.00 450,057.99
Model with frequency of touch − 225,019.20 450,242.40 450,529.39
instead of dwell times
Model without touch gestures − 230,059.90 460,193.80 460,297.90
the power of touch-stream data, especially the dwell time of
informational touch features, in explaining consumer app
browsing behaviors. Next, we describe the results from the
main model (Table 3).
Level 1: App Stay Vs. Exit Decision Results
Within-session Experience
Our model results reveal strong explanatory power of the
three informational touch gestures on app stay or exit decisions
over and above cumulative session time, which supports our
conjecture. A larger number of unique informational touch
gestures up to time t − 1 (− 0.326) leads to a greater likelihood
of exiting the app, suggesting that if a consumer uses all three
touch features, he or she would be more likely to exit.
However, a navigational touch, which does not gather
additional information, is not predictive.
The effects of informational touch gestures on app engage-
ment diverge. Longer cumulative dwell time of product-view up
to time t − 1 (− 0.606) will lead to a greater likelihood of exiting
the app; whereas a longer dwell time of zoom-page has the
opposite effect (0.437). This pattern confirms our prediction. The
timing of the first-time use of a touch feature during a session
matters too. If a consumer starts to use zoom-page earlier in a
session, i.e., during the 1st or 2nd quarters, he or she is more
likely to stay with the app (0.371 for the 1st quarter and 0.115 for
the 2nd quarter); however, adopting this touch feature late will
not exert significant impact on app engagement. Conversely,
adopting product-view during the 1st quarter encourages
consumers to exit the app early (− 0.159).
Across-session Experience
Across-session experience affects the app stay or exit decision
in a different fashion than within-session experience. When
consumers accumulate greater experience with the target app
(cumulative time), their likelihood of staying with the app in the
current session will be higher (0.217). Cumulative informational
touch gestures in prior sessions, instead of navigational ones,
also positively influence the stay likelihood (0.192).
Control Variables
As for the control variables, a minor app upgrade performed
since the last visit will increase the stay likelihood (0.285); yet a
major upgrade has the opposite effect (− 0.135), suggesting that
the consumer experience has been disrupted materially. It is
likely that after a major upgrade, consumers are merely
checking out the new features rather than performing a regular
browsing session. The weekend and holiday indicators are
found to be less predictive of app stay or exit decisions.
49
Level 2: Touch Gesture and New Page Decisions Results
Next, we look at the factors driving the choices among
product-view, zoom-page, open-webpage, and going to a new
page (baseline) at the second level, conditional on continued
visit on the app (results are presented in the second half of
Table 3).
Within-session Experience
First, the baseline (intercept) of choosing which of the three
touch gestures to use varies, with the use of zoom-page being the
most likely (1.432), followed by product-view (0.804), and open-
webpage (− 0.369) (move to a new page as baseline 0). This result
is in line with our prediction. Such differences may be attributed to
the fact that open-webpage creates a non-native browsing
experience, which incurs additional cognitive or time costs and
interrupts the browsing experience. Another possible explanation
is that these touch features represent different stages in a purchase
(conversion) funnel. The use of open-webpage, which represents
the lower end of the funnel or strongest intention to purchase, is
less frequent than zoom-page, which represents the higher end of
the funnel, or initial interest. Our survey may shed some light on
these two alternative explanations. The participants in our survey
indicated that zoom-page was the easiest to use and open-
webpage was the most difficult. Thus, we believe that the native
vs. non-native experience may have played a significant role.
Second, the drivers of the use of these touch features vary.
Greater time spent in a session will lead to the increased
likelihood of using zoom-page (0.744), while the effect is much
smaller for product-view (0.392) and become negative for
open-webpage (− 0.304). The cumulative navigational swipes
only predict zoom-page (0.462). The cumulative number of
unique informational touch gestures only increase the usage
rate of product-view (0.423) and open-webpage (− 0.111).
Third, as predicted, these touch gestures exhibit strong
synergy and antergy. A longer cumulative dwell time of
product-view significantly reduces the likelihood of using
zoom-page (− 0.220). The dwell time of open-webpage exerts
an even stronger negative effect on zoom-page (− 0.388). The
probability of using product-view rises when the cumulative
time spent in open-webpage increased (0.802). If a consumer
dwells longer in page-view (0.814) prior to time t, the
likelihood of using open-webpage at time t increases; however,
the cumulative dwell time of zoom-page (− 0.329) has an
opposite effect on the use of open-webpage.
The timing of the first-use of one touch feature influence the
use of the others as well. An early use of open-webpage will
reduce the usage rate of zoom-page (1st quarter OW: − 0.160).
If a consumer start to use zoom-page and product-view during
the 1st quarter of a session, he or she is more likely to use
product-view in that session (0.162 and 0.108, respectively);
however, the usage rate of product-view will be undermined if
zoom-page is adopted later in a session (3rd quarter, − 0.095).
As for open-webpage, its use likelihood will improve if
product-view is adopted in the 1st and 2nd quarters (0.160
and 0.116, respectively); early adoption of open-webpage (1st
quarter 0.072) also exerts a positive impact.
50 S.W. Shi, K. Kalyanam / Journal of Interactive Marketing 44 (2018) 43–59

Table 3
Main model results.
Level 1: Stay–exit decision Variables Mean Posterior 2.50% Posterior 97.50%
Intercept 0.382 0.330 0.435
Within-session experience CumuTimeSession (log) − 0.309 − 0.368 − 0.250
CumuSwipeSession (log) − 0.033 − 0.106 0.041
CumuTypeTouchSession − 0.326 − 0.530 − 0.123
CumuZPTimeSession (log) 0.437 0.305 0.570
CumuPVTimeSession (log) − 0.606 − 0.679 − 0.533
CumuOWTimeSession (log) − 0.079 − 0.347 0.188
1stQuarterObserveFirstZC 0.371 0.316 0.427
2ndQuarterObserveFirstZC 0.115 0.099 0.132
3rdQuarterObserveFirstZC − 0.016 − 0.091 0.059
1stQuarterObserveFirstPV − 0.159 − 0.231 − 0.088
2ndQuarterObserveFirstPV − 0.106 − 0.285 0.073
3rdQuarterObserveFirstPV 0.060 − 0.078 0.198
1stQuarterObserveFirstOW 0.091 − 0.039 0.221
2ndQuarterObserveFirstOW − 0.034 − 0.107 0.039
3rdQuarterObserveFirstOW − 0.009 − 0.042 0.024
Across-session experience CumuTimeApp (log) 0.217 0.142 0.292
CumuSwipeApp (log) − 0.038 − 0.109 0.032
CumuTouchApp (log) 0.192 0.176 0.208
Control variables Minor Upgrade 0.285 0.255 0.316
Major Upgrade − 0.135 − 0.187 −0.084
Weekend 0.284 − 0.032 0.601
Peak (holiday) 0.061 0.047 0.075
Inclusive value Parameter for IV 0.630 0.480 0.779

Level 2: Variables Zoom-page Product-view Open-webpage

Touch
Mean Posterior Posterior Mean Posterior Posterior Mean Posterior Posterior
choice
2.50% 97.50% 2.50% 97.50% 2.50% 97.50%
Intercept 1.143 0.972 1.313 0.804 0.732 0.877 − 0.369 − 0.474 − 0.264
Within- experience
session CumuTimeSession (log) 0.744 0.713 0.776 0.392 0.355 0.429 − 0.304 − 0.337 − 0.271
CumuSwipeSession (log) 0.462 0.259 0.665 0.025 − 0.132 0.183 0.073 − 0.183 0.328
CumuTypeTouchSession − 0.052 − 0.127 0.022 0.423 0.355 0.491 − 0.111 − 0.167 − 0.055
CumuZPTimeSession (log) 0.372 0.184 0.560 − 0.082 − 0.347 0.184 − 0.329 − 0.376 − 0.283
CumuPVTimeSession (log) −0.220 − 0.294 − 0.147 0.173 − 0.072 0.418 0.814 0.610 1.017
CumuOWTimeSession −0.388 − 0.472 − 0.305 0.802 0.392 1.212 0.157 − 0.391 0.705
(log)
1stQuarterObserveFirstZC − 0.308 − 0.648 0.032 0.162 0.147 0.177 − 0.055 − 0.130 0.019
2ndQuarterObserveFirstZC − 0.146 − 0.400 0.107 0.044 − 0.049 0.137 − 0.031 − 0.063 0.001
3rdQuarterObserveFirstZC − 0.200 − 0.667 0.266 −0.095 − 0.150 − 0.039 0.008 − 0.070 0.086
1stQuarterObserveFirstPV 0.207 − 0.520 0.933 0.108 0.070 0.147 0.160 0.038 0.281
2ndQuarterObserveFirstPV 0.036 − 0.076 0.148 0.048 − 0.021 0.118 0.116 0.065 0.167
3rdQuarterObserveFirstPV − 0.188 − 0.399 0.023 − 0.064 − 0.168 0.040 − 0.184 − 0.583 0.215
1stQuarterObserveFirstOW −0.160 − 0.232 − 0.088 0.152 − 0.051 0.355 0.072 0.055 0.090
2ndQuarterObserveFirstOW 0.085 − 0.088 0.259 − 0.037 − 0.171 0.096 − 0.083 − 0.216 0.049
3rdQuarterObserveFirstOW 0.121 − 0.025 0.266 0.074 − 0.031 0.179 0.063 − 0.128 0.254
Across- experience
session CumuTimeApp (log) − 0.209 − 0.223 − 0.196 0.129 0.124 0.134 0.123 0.048 0.199
CumuSwipeApp (log) − 0.088 − 0.228 0.052 − 0.012 − 0.088 0.065 0.027 − 0.059 0.112
CumuTouchApp (log) 0.053 − 0.017 0.122 0.417 0.190 0.644 0.176 0.167 0.184
Page-level experience
ZPDwellTimePage (log) 0.183 0.172 0.193 − 0.266 − 1.165 0.633 0.033 − 0.094 0.160
PVDwellTimePage (log) −0.045 − 0.055 − 0.036 −0.228 − 0.349 − 0.107 − 0.040 − 0.219 0.139
OWDwellTimePage (log) − 0.009 − 0.176 0.158 −0.075 − 0.090 − 0.060 − 0.186 − 0.507 0.134
Control Minor upgrade − 0.059 − 0.178 0.061 0.306 − 0.121 0.733 0.124 0.055 0.193
variables Major upgrade 0.081 0.049 0.113 − 0.042 − 0.128 0.044 0.053 − 0.048 0.154
Weekend − 0.062 − 0.139 0.016 0.110 0.085 0.134 0.075 − 0.049 0.199
Peak 0.021 − 0.053 0.095 − 0.158 − 0.289 − 0.027 0.063 − 0.118 0.244
Note: The bold font indicates that the 95% credible interval does not cover zero.
 S.W. Shi, K. Kalyanam / Journal of Interactive Marketing 44 (2018) 43–59
Across-session Experience
As for the experience level measured across sessions, we find
that the cumulative time spent in the app lifts the baseline of
product-view (0.129) and open-webpage use (0.123), yet
decreases the baseline of zoom-page use (− 0.209); the use
frequency of informational touch gestures in the app increases the
baseline for product-view (0.417) and open-webpage (0.176).
Page-level Dynamics
We find that the use of zoom-page is more likely if a
consumer has dwelled longer with zoom-page on the same page
(0.183). This is consistent with the results of within-session
experience, in which longer cumulative dwell time of zoom-page
increases the likelihood of using that touch gesture in the future
(0.372). One potential explanation is that zoom-page is subject to
repetitive use to allow the consumer to see multiple aspects of
the same product. On the other hand, the time spent in page-view
on the same page has a negative effect on the likelihood of
zoom-page use (− 0.045). Longer dwell time on page-view and
open-webpage on the same page also reduces the use likelihood
of product-view (− 0.228 and − 0.075, respectively).
Control Variables
We find that the performance of a major upgrade since the
consumer's last visit significantly increases the likelihood of using
zoom-page (0.081). A minor upgrade encourages the use of open-
webpage (0.124). It is possible that whenever the app introduces
“major features,” consumers hesitate to make purchases (through
open-webpage) and instead waited for that version of the app to
stabilize. A minor upgrade or bug fix gives consumers more
confidence for purchasing on the platform and thus open-webpage
becomes more likely. The baseline of product-view use will be
higher when the session happens on a weekend (0.110).
Managerial Implications
Touch, as an inseparable design feature in mobile apps,
presents new challenges and opportunities. This paper empir-
ically examines touch behavior with real-life mobile shopping
data. In this section, we discuss the managerial implications of
our findings. We start with a summary of these implications.
1. Regarding the impact of touch gestures on the app stay
likelihood:
• Our results demonstrate the strong explanatory power of
informational touch gestures, especially their dwell time,
in consumer browsing decisions; whereas the use of
navigational touch gestures does not exert significant
impact on app stay likelihood. Thus, it is crucial for
marketers to closely monitor informational touch gestures
when assessing consumer browsing decisions.
• A longer dwell time and early use of zoom-page within a
session improve app stay likelihood. This effect exhibits
considerable heterogeneity across consumers. These
results have implications for potential interventions
regarding the use of zoom-page, especially in terms (a)
51
the timing of the intervention and (b) the potential for
individual-level targeting;
• The use of certain informational touch features discour-
ages app stay and, thus, should be inserted with caution.
2. Regarding the patterns of touch feature use:
• We observe strong synergy and antergy among the
informational touch gestures. This suggests that managers
can (a) infer stage of the shopping process based on touch
gestures (segmentation) and (b) guide consumers through
the purchase funnel by promoting the use of certain touch
features;
• The open-webpage feature, which creates a non-native
experience, has the lowest baseline usage rate. Given the
importance of this feature in transactions, marketers can
(a) promote its use based on the synergy among touch
features and (b) smooth the non-native experience through
app design.
Next, we discuss each of these implications in more detail.
The Impact of Touch Gestures on App Stay Likelihood
The Explanatory Power of Touch Gestures
It is plausible that the strong explanatory power of dwell
time of informational touch gestures for app stay likelihood is
related to the limited real estate on mobile platforms. Because
abbreviated information is presented on the default page in
apps, informational touch gestures are applied to retrieve
detailed, inner-layer information that is not presented by
default. The active use of informational touch requires extra
cognitive and physical effort (e.g., Heo and Lee 2011; Kolb,
Rudner, and Reichert 2012; Wilson and Kinlan 2013) and, thus,
might be associated with stronger purchase intent compared
with navigational touch. Therefore, it is beneficial for marketers
to collect individual-level data as opposed to aggregate data on
the use of these informational touch gestures, especially dwell
time, when assessing consumer browsing decisions.
Intervention with Zoom-page: Timing
As consumer browsing time accumulates, their likelihood of
staying with the app diminishes. The use of zoom-page,
especially when it is used early in a session, attenuates this
effect. This empirical finding underscores the importance of
engaging consumers early when designing catalogs: inserting
an eye-catching product early in a catalog to entice zooming-in
can improve stay likelihood. We illustrate this practical benefit
of early engagement with zoom-page through a simulation.
Using the estimated parameters for all consumers, we first
calculate the stay probabilities at each time point when zoom-
page is not used (baseline). Then, we simulate the new stay
probabilities in which app developers intervene, for example,
with an eye-catching product that prompt consumers to use
zoom-page during the 1st, 2nd, 3rd, and 4th quarters of a
browsing session. Lastly, we calculate incremental stay
probabilities by subtracting the baseline stay probabilities
from the new stay probabilities for each consumer under each
intervention. Fig. 2.1 illustrates the average incremental stay
52 S.W. Shi, K. Kalyanam / Journal of Interactive Marketing 44 (2018) 43–59

probabilities when marketers intervene in each of the four
quarters. Prompting the use of zoom-page increases the app
stay probabilities materially compared to the situation when
zoom-page is not used. Furthermore, the earlier the marketer
encourages the zoom-page gesture, the higher the average
incremental stay probabilities (e.g., intervening at 1st quarter
results in an average increase of 0.089 for stay probabilities
over time, while the number reduces to 0.001 when intervening
in the 4th quarter); and this incremental effect does not wear off
for a long period of time.
Intervention with Zoom-page: Individual-level Targeting
It is worth noting that the effects of the intervention vary
substantially across consumers, as shown by the large standard
deviation relative to the mean value of the incremental stay
probabilities (the Table under Fig. 2.1). Fig. 2.2 illustrates the
incremental stay probabilities for two sample consumers using
individual-level parameters. As the figure shows, their
responses to the intervention with zoom-page at the 1st quarter
for ID1 and ID2 are substantially different: the average
incremental stay probability for ID1 is 0.186 whereas that for
ID2 is 0.064. Markers can apply our framework to quantify the
impact of such interventions on consumer browsing decisions
at the individual level.
Note that the targeted intervention is of greater importance
for certain interventions, as revealed by the dispersion of the
incremental stay probabilities across the population. In Fig. 2.3,
we plot the kernel densities of the incremental stay probabilities
across all consumers with the zoom-page intervention in
different quarters. An intervention in the 4th quarter creates
the highest level of concentration in the distribution of
incremental stay probabilities, whereas an intervention in the
1st quarter is effective for a wider range of consumers
(corresponding to the Gini indices in Table under Fig. 2.1).
Percentage-wise, the top 10% of consumers make up 27.54% of
the improvement (in stay probabilities) for an intervention in

Average Incremental Stay Probabilities When Promoting Zoom-Page

The average incremental stay probabilities when zoom-page (ZP) is prompt to be adopted in Q1
(1st quarter, black line), Q2 (2nd quarter, gray dotted line), Q3 (3rd quarter, gray line), and Q4 (4th
quarter, black dash line), compared to the stay probabilities when zoom-page is not adopted. To
simplify, we set the time point for Q1, Q2, Q3, and Q4 intervention to be 3, 8, 13, and 18.

0.28

0.24
Incremental Stay Probabilities

0.2

0.16

0.12

0.08

0.04

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Time

Intervene at Q1 Intervene at Q2
Intervene at Q3 Intervene at Q4

Simulation is based on the following parameters of the population estimates (level 1 model): baseline;
CumuTimeSession; CumuTypeTouchSession; 1stQuarterObserveFirstZC; 2ndQuarterObserveFirstZC;
3rdQuarterObserveFirstZC; CumuZPTimeSession. To simplify, we set time to be from 1 to 20 and use
9.64 seconds as average dwell time of ZP.

Summary Statistics of Incremental Stay Probabilities across Time Points

Incremental Incremental Incremental Incremental
Prob. when Prob. when Prob. when Prob. when
Intervene at Q1 Intervene at Q2 Intervene at Q3 Intervene at Q4
Mean* 0.089 0.023 0.004 0.001
SD 0.026 0.011 0.003 0.001
Gini Index for
Incremental 0.167 0.265 0.357 0.402
Prob.
Note: * the mean, SD, and Gini index are calculated based on the population estimates.

Fig. 2. Illustration of the application of model results. 2.1. Average incremental stay probabilities when promoting zoom-page. 2.2. Heterogeneity of incremental stay
probabilities when promoting zoom-page in the first quarter. 2.3. Dispersion of incremental stay probabilities across the population when intervene with zoom-page
(kernel density). 2.4. Intervention to encourage the use of product-view. 2.5. Illustration of complimentary touch feature.
 S.W. Shi, K. Kalyanam / Journal of Interactive Marketing 44 (2018) 43–59 53

Heterogeneity of Incremental Stay Probabilities When Promoting Zoom-Page

in the First Quarter
To simplify, we set the time point for Q1 intervention (1st quarter) to be 3.
0.35

0.3

Incremental Stay Probabilities 0.25

0.2

0.15

0.1

0.05

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Time

ID1_intervene with ZP at Q1 ID2_intervene with ZP at Q1

Simulation based on the following parameters:

ID1 ID2
Intercept 0.396 Intercept 0.378
CumuTimeSession -0.255 CumuTimeSession -0.33
CumuTypeTouchSession -0.292 CumuTypeTouchSession -0.341
1stQuarterObserveFirstZC 0.422 1stQuarterObserveFirstZC 0.329
CumuZPTimeSession 0.55 CumuZPTimeSession 0.338
1stQuarterObserveFirstPV -0.101 1stQuarterObserveFirstPV -0.219
CumuPVTimeSession -0.553 CumuPVTimeSession -0.647

Dispersion of Incremental Stay Probabilities across the Population When

Intervene with Zoom-page (Kernel Density)
54 S.W. Shi, K. Kalyanam / Journal of Interactive Marketing 44 (2018) 43–59

Intervention to Encourage the Use of Product-View

The probabilities of adopting product-view: baseline (no intervention, black line); when zoom-page
(ZP) is prompt to be adopted in Q1 (1st quarter, gray line); and when product-view (PV) is prompt
to be adopted in Q1 (1st quarter, black dash line). To simplify, we set the time point for Q1
intervention (1st quarter) to be 3.

0.23

0.21

0.19
Probabilty of Product View

0.17

0.15

0.13

0.11

0.09
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Time

Baseline PV Intro ZP in Q1 Intro PV in Q1

Simulation is based on the following parameters for three features in level 2 model: ZP: baseline,
CumuTimeSession; 1stQuarterObserveFirstOW; PV: baseline, CumuTimeSession;
1stQuarterObserveFirstZC; 1stQuarterObserveFirstPV; OW: baseline, CumuTimeSession;
1stQuarterObserveFirstPV, and 1stQuarterObserveFirstOW.

Baseline Intro ZP in Q1 Intro PV in Q1

Avg. Prob. of Adopting PV 0.132 0.148 0.143
Perc. Increase from Baseline 12.63% 8.23%

Illustration of Complimentary Touch Feature

Open-Webpage Conditional on Product-View (Black line)
Open-Webpage Baseline (No Product-View Usage, Gray Line)

1
Probilities of Open-webpage

0.8

0.6

0.4

0.2

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Frequency of Usage of Product_View

open-webpage|product-view open-webpage baseline

Simulation is based on the following parameters in level 2 model: OW baseline; OW-

CumuPVtime; ZP baseline; ZP-CumuPVtime; PV baseline.
 S.W. Shi, K. Kalyanam / Journal of Interactive Marketing 44 (2018) 43–59
the 4th quarter, whereas the top 10% of consumers only make
up 15.51% of the improvement for an intervention in the 1st
quarter. Therefore, when resources are limited, it is of greater
importance to release a targeted intervention when such an
intervention happens later in a browsing session, compared
with that happens earlier in a browsing session.
Touch Features to be Inserted with Caution
At first glance, it would seem that providing various kinds of
touch features might create rich tactile feedback for consumers
and elicit more interest in apps. However, app developers
should balance such benefits with potential downsides to
consumer engagement levels, in addition to the extra effort of
recoding apps, reconfiguring webpages, and testing before
introducing touch features. Our results show that use of
informational touch gestures has a negative impact on stay
probability: the longer dwell time on product-view and more
(unique) informational touch gestures used in a session
(CumuTypeTouchSession) lower the probability of staying. It
is possible that this finding is related to the physical effort
involved in browsing from an ergonomic perspective (e.g., Heo
and Lee 2011; Kolb, Rudner, and Reichert 2012). It is also
possible that consumers quickly deplete their cognitive
resources after scanning multiple product pages (zoom-page),
performing a large number of detailed product examinations
(product-view), or encountering significant friction in the
transitions between multiple windows or loading embedded
links (open-webpage). Information saturation could have also
played a role: as consumers retrieve enough detail about the
product, they may not need to continue browsing. With the
current dataset, we are not able to test between explanations
and, thus, leave it to future research. Nevertheless, app
developers can use our results to understand which touch
features extend app stay likelihood and to prioritize touch
feature design accordingly.
The Patterns of Touch Feature Use
Segmentation Based on Touch Gestures
We find a strong synergy between the use of open-webpage
and product-view, and a negative impact of cumulative dwell
time of these two gestures on the use of zoom-page, at both the
session and page level. These patterns are consistent with
different search goals and purchase intents underlying these
touch features. The use of product-view and open-webpage
suggests that consumers are purposefully expending extra effort
to perform directed, focused searches on inner-layer product
information that is not presented by default. This reflects an
increased level of app engagement and perhaps a later stage of
the purchase process. However, the use of zoom-page merely
enlarges the images and is consistent with initial interest and
perhaps an earlier stage of the purchase process (exploration).
We note that there is a difference between product-view and
open-webpage, as the cumulative use of unique informational
touch in a session encourages the use of product-view yet
discourages the use of open-webpage. Product-view represents
a knowledge-building process (Moe 2003) whereby consumers
55
accumulate knowledge about products; therefore, they are
conducting information retrieval through informational touch
gestures. Open-webpage, on the other hand, represents a direct-
buy process (Moe 2003) whereby consumers are making
immediate purchases rather than collecting information through
informational touch gestures.
By closely monitoring informational touch gestures, managers
can identify different consumer segments on a mobile shopping
platform. A consumer who uses zoom-page more frequently than
product-view or open-webpage may be in an earlier stage of the
purchase process; such a consumer may be more receptive to
recommendations or ads that help with knowledge-building;
whereas a consumer who uses open-webpage more frequently is
likely to be at a later stage of the purchase process and may be
more responsive to coupons to finalize their transactions.
Furthermore, longer cumulative time spent in the app across
sessions positively affects the use of product-view and open-
webpage but has a negative impact on the use of zoom-page.
This implies that consumers increasingly rely on the app as a
shopping platform for detailed, additional information retrieval
and transactions (with product-view and open-webpage), rather
than only use the app as a hedonic browsing platform (with
zoom-page), as they accrue experience across sessions with the
app. The cumulative time spent in the app might be a useful
proxy for targeting hedonic browsers versus those with stronger
purchase intent.
Purchase Funnel and Intervention with Touch Features
Our estimates show that the timing of the first-time use of
touch features during a session alters the likelihood of use of
these features later in that session. Specifically, using zoom-page
and product-view earlier result in a greater likelihood of using
product-view in that session, with the first-time use of zoom-
page exerting a larger impact. Getting consumers to consider
product information in detail (with product-view) is a crucial
objective in marketing because it signals a later stage of the
purchase funnel. Our results suggest a way for managers to
enable this. In Fig. 2.4, we illustrate this managerial implication
through a simulation in which app developers prompt consumers
to use zoom-page or product-view during the 1st quarter of their
browsing sessions. We find that the most effective way to drive
consumers to use a touch feature that explores inner-layer
product details (with product-view), is to prompt them to use
zoom-page early during a session (on average, it increases the
usage rate of product-view by 12.63%); introducing product-
view during the 1st quarter is also beneficial, yet the effect is
slightly smaller (on average, it increases the usage rate of
product-view by 8.22%). Therefore, app developers can
encourage consumers to move into a later purchase stage and
make more frequent use of product-view by engaging them early
with exploratory touch features such as zoom-page. Introducing
a catchy or visually engaging product earlier in the catalog might
be helpful in eliciting the use of zoom-page early in a session.
Transactions and the Use of Open-webpage
Open-webpage is a desirable feature for shopping apps
because it enables consumers to proceed further in the
56 S.W. Shi, K. Kalyanam / Journal of Interactive Marketing 44 (2018) 43–59

transaction process and contributes to affiliate commissions for
app developers. Our results suggest an opportunity for
marketers to promote the use of open-webpage based on the
synergy among touch gestures. We observe strong feature
dependence between open-webpage and product-view, suggest-
ing that perhaps a similar underlying variable of purchase intent
drives these two variables. Indeed, when we compare the
conditional use probability of open-webpage, given that
product-view is used versus not used (Fig. 2.5), the use of
open-webpage became much more likely when a consumer
starts to use product-view, compared to when the consumer
does not use product-view at all.
However, among all three touch features, open-webpage has
the lowest likelihood of being used. We suspect that this is
related to its non-native implementation. This touch feature
either connects to an embedded webpage link within the app or
directs consumers out of the current app to another window.
However, most mobile platforms do not currently support
switching seamlessly among multiple windows or loading
embedded pages (e.g., Lawrence, Pernici, and Krogstie 2004),
which means that consumers must either exit the current task
completely to perform another task, or experience a significant
delay in page loading within the current app. It is well known
that page load delays are detrimental to user experience (e.g.,
Galletta et al. 2004). This is even more acute for mobile devices
than for PCs (DoubleClick 2016). The mental and physical
costs of open-webpage are likely to be the highest. Indeed, the
participants in our survey indicated that this feature was the
most difficult to use. It is advisable for app developers to seek
ways to ease the non-native experience. Facebook's instant
article program took a first step in this direction by allowing
media companies to publish their content directly on Facebook
user feeds rather than linking back to media on the companies'
websites. This shortens the loading time greatly (10 times faster
than an embedded link) and improves the consumers' non-
native experiences.33
Limitations and Future Research
Our study contains some limitations. We performed an
exploratory empirical study of touch features based on
observational data; other researchers might expand on this
data using field experiments to test the location, intensity, and
variety of touch feature designs. With the development of
technology, new tactile feedback is sure to be introduced to
mobile devices, from simple clicks to complex vibrotactile
patterns. Thus, broadening the scope of touch features on
mobile devices is a promising avenue for future research. We
believe that examining touch features on mobile platforms and
their impact on consumer browsing behaviors in this emerging

e-commerce channel is of great importance to both marketing researchers and mobile commerce practitioners, and we hope that our
research elicits more interest in this area.

Appendix 1. Illustration of Three Touch Features

a. Touch to zoom-in on a (catalog) page (illustration with the app “the Coffee Table”)

3
All Washington Post Stories to Go to Facebook, by Lukas I. Alpert, Wall Street Journal, Sept. 22, 2015, http://www.wsj.com/articles/all-washington-post-stories-
to-go-to-facebook-1442961105, retrieved on Sept. 15, 2015 from https://instantarticles.fb.com/.
 S.W. Shi, K. Kalyanam / Journal of Interactive Marketing 44 (2018) 43–59 57

b. Touch to view product (Illustration with the app, “Shop Spree”/WiiU game touch point)

c. Touch to open product website within the App (illustration with the app, “Polyvore”, “Buy AT UGGAUSTRIA.COM” touch
point)
58 S.W. Shi, K. Kalyanam / Journal of Interactive Marketing 44 (2018) 43–59
Appendix 2. Survey to Understand Touch Usage
The main intention was to validate the basic usage pattern of
three touch features and to discover whether the usage is driven by
the increased experience or by learning a new functionality. There
were three parts of questions for each touch feature. The first part
asked participants to rate the difficulty level of using a certain
touch feature. The second part asked participants whether,
through multiple uses of the app, they discovered new functions
of the target touch feature, or became more familiar with that
feature. The third part asked participants if they started to use
certain touch feature more often over time, and if so, was it
because they discovered new functions of this touch feature or
became more familiar with the touch feature. Lastly, we asked
participants to rank the ease of use of the three touch features (with
1 being the easiest, and 3, the hardiest). We recruited 46 students
from a university on West Coast of the United States for this
survey. Participants had never used the target app before and were
asked to download the target app and play with it over two weeks
before completing the survey. We ended up with 43 usable
samples.
We find that all three touch features are relatively easy to use
(2.186 for zoom-page, 2.326 for product-view, and 2.419 for
open-webpage on a seven-point Liker scale, with higher
numbers indicating more difficult to use). This pattern
coincides with consumers' ranking of the ease of use for the
three touch features, with open-webpage being the most
difficult one, followed by product-view and zoom-page
(average ranking: 2.25, 1.84, and 1.74, respectively; p b  .
0.000). Participants indicate that they become more familiar
with touch features through multiple uses of the app, more so
than they discover new functions of the touch features (see
Fig. A2.1). Out of the 43 participants, 33 report using the
zoom-page touch feature more often over time, while the
corresponding number for product-view and open-webpage is
28 and 20, suggesting a positive relationship between perceived
ease of use and usage frequency. For those consumers who use
zoom-page more often over time, they indicate that the
familiarity with this touch feature is a stronger driving force
compared with discovering new functionality (p b 0.000) of
this touch feature. This also holds for product-view (Fig. A2.2).
Fig. A2.1. Responses to “Through multiple usage of the app, I have discovered
the new functions of this touch gesture” versus “I have become more familiar
with this touch gesture.”
7
1
Zoom_Page Product_View Open_Webpage
Discover new function Become more familiar
Fig. A2.2. Responses to “I started to use the touch gesture more often over time
(if so) because have discovered the new functions of this touch gesture” versus
“I have become more familiar with this touch gesture.”
7
6
5
4
3
2
1
Zoom_Page Product_View Open_Webpage
Discover new function Become more familiar
References
Brasel, S. Adam and James Gips (2014), “Tablets, Touchscreens, and
Touchpads: How Varying Touch Interfaces Trigger Psychological Owner-
ship and Endowment,” Journal of Consumer Psychology, 24, 2, 226–33.
Coyle, James R. and Esther Thorson (2001), “The Effects of Progressive Levels
of Interactivity and Vividness in Web Marketing Sites,” Journal of
Advertising, 30, 3, 65–77.
DoubleClick (2016), “The Need for Mobile Speed: How Mobile Latency
Impacts Publisher Revenue,” by Google, Retrieved on Feb 20th, 2017,
from: https://www.doubleclickbygoogle.com/articles/mobile-speed-matters/
Forrester Research (2014), “US Mobile and Tablet Commerce to Top $293b by
2018; Total Ecommerce to Hit $414b,” Retrieved on Dec 5th, 2015, from
https://www.forrester.com/US+Mobile+And+Tablet+Commerce+To+Top
+293B+by+2018+Total+eCommerce+To+Hit+414B/-/E-PRE7004.
 S.W. Shi, K. Kalyanam / Journal of Interactive Marketing 44 (2018) 43–59
Galletta, Dennis F., Raymond Henry, Scott McCoy, and Peter Polak (2004),
“Web Site Delays: How Tolerant are Users?” Journal of the Association for
Information Systems, 5, 1,1–28.
He, Jibo and Jason S. McCarley (2010), “Executive Working Memory Load
Does Not Compromise Perceptual Processing During Visual Search:
Evidence From Additive Factors Analysis,” Attention, Perception, &
Psychophysics, 72, 2, 308–16.
Heo, Seongkook and Geehyuk Lee (2011), “Force Gestures: Augmenting
Touch Screen Gestures with Normal and Tangential Forces,” Proceedings
of the 24th Annual ACM Symposium on User Interface Software and
Technology, ACM, 621–6.
Klatzky, Roberta L. and Susan J. Lederman (1992), “Stages of Manual
Exploration in Haptic Object Identification,” Perception & Psychophysics,
52, 6, 661–70.
Koenigstorfer, Joerg and Andrea Groeppel-Klein (2012), “Consumer Accep-
tance of the Mobile Internet,” Marketing Letters, 23, 4, 917–28.
Kolb, Jens, Benjamin Rudner, and Manfred Reichert (2012), “Towards
Gesture-based Process Modeling on Multi-touch Devices,” Advanced
Information Systems Engineering Workshops, Springer Berlin Heidelberg.
Laroche, Michel, Jasmin Bergeron, and Christine Goutaland (2001), “A Three-
dimensional Scale of Intangibility,” Journal of Service Research, 4, 1,
26–38.
Lawrence, Elaine, Pernici, Barbara, Krogstie, John, editors. Mobile Information
Systems, , Springer Science & Business Media, 15–7.
Leppäniemi, Matti and Heikki Karjaluoto (2008), “Mobile Marketing: From
Marketing Strategy to Mobile Marketing Campaign Implementation,”
International Journal of Mobile Marketing, 3, 1,50–61.
Moe, Wendy (2003), “Buying, Searching, or Browsing: Differentiating
Between Online Shoppers Using In-store Navigational Clickstream,”
Journal of Consumer Psychology, 13, 1, 29–39.
Peck, Joann and Terry L. Childers (2003a), “To Have and To Hold: The
Influence of Haptic Information on Product Judgments,” Journal of
Marketing, 67, 2, 35–48.
——— and ——— (2003b), “Individual Differences in Haptic Information
Processing: The “Need for Touch” Scale,” Journal of Consumer Research,
30, 3, 430–42.
——— and Suzanne B. Shu (2009), “The Effect of Mere Touch on Perceived
Ownership,” Journal of Consumer Research, 36, 3, 434–47.
Peterson, Matthew S. and Jason H. Wong (2008), “Were You Paying Attention
to Where You Looked? The Role of Executive Working Memory in Visual
Search,” Psychonomic Bulletin & Review, 15, 2, 372–7.
Powers, Daniel and Yu Xie (2008), Statistical Methods for Categorical Data
Analysis. Emerald Group Publishing.
Salesforce (2014), “2014 Mobile Behavior Report,” retrieved on May 11, 2017,
from https://www.marketingcloud.com/sites/exacttarget/files/deliverables/
etmc-2014mobilebehaviorreport.pdf.
59
Shankar, Venkatesh and Sridhar Balasubramanian (2008), “Mobile Marketing:
Synthesis and Prognosis,” Journal of Interactive Marketing, 23, 2, 118–29.
———, Alladi Venkatesh, Charles Hofacker, and Prasad Naik (2010), “Mobile
Marketing in the Retailing Environment: Current Insights and Future
Research Avenues,” Journal of Interactive Marketing, 24, 2, 111–20.
Smith, Robert E. and William R. Swinyard (1983), “Attitude–Behavior
Consistency: The Impact of Product Trial Versus Advertising,” Journal of
Marketing Research, 20, 3, 257–67.
Tokárová, Lucia and Melius Weideman (2013), “Understanding the Process of
Learning Touch-screen Mobile Applications,” Proceedings of the 31st ACM
International Conference on Design of Communication, ACM, 157–64.
Train, Kenneth (2009), Discrete Choice Methods with Simulation. Cambridge
University Press.
Ullrich, Chris (2005), Haptic Phone Gaming Study. Immersion Corporation.
Underwood, Geoffrey, Lorraine Jebbett, and Katharine Roberts (2004),
“Inspecting Pictures for Information to Verify a Sentence: Eye Movements
in General Encoding and in Focused Search,” The Quarterly Journal of
Experimental Psychology, 57, 165–82.
Van Doorn, Jenny, Katherine N. Lemon, Vikas Mittal, Stephen Naß, Doreén
Pick, Peter Pirner, and Peter Verhoef (2010), “Consumer Engagement
Behavior: Theoretical Foundations and Research Directions,” Journal of
Service Research, 13, 3, 253–66.
Verhagen, Tibert, Charlotte Vonkeman, Frans Feldberg, and Plon Verhagen
(2013), “Making Online Products More Tangible and Likeable: The Role of
Local Presence as Product Presentation Mechanism,” Research Memoran-
dum, 3.
Villamor, Craig, Dan Willis, and Luke Wroblewski (2010), “Touch Gestures
Reference Guide,” Retrieved on Jun 20th, 2017, from: http://static.lukew.
com/TouchGestureGuide.pdf.
Wilson, Chris and Paul Kinlan (2013), “Touch and Mouse: Together Again for
the First Time,” Retrieved on May 11, 2017, from https://www.html5rocks.
com/en/mobile/touchandmouse/.
Savannah Wei Shi is an Associate Professor of Marketing at the Leavey
School of Business, Santa Clara University. Her research focuses on e-
commerce, internet marketing, and dynamics in consumer decision-making.
Kirthi Kalyanam is the J.C. Penney Research Professor and Director of the
Retail Management Institute at the Leavey School of Business at Santa Clara
University. He has also served as Faculty Director of the Executive MBA
program, a visiting Professor at the Graduate School of Business at Stanford
University, a guest faculty in the Stanford Executive MMP program, Senior
Vice President and Chief Marketing Officer of SpinCircuit Inc., a provider of
supply chain integration services to the electronics industry. His research and
expertise are in retailing, internet & multi-channel marketing, quantitative
marketing and the intersection of these areas.